llvm-project/llvm/test/Transforms/MemCpyOpt/memcpy.ll

212 lines
6.8 KiB
LLVM
Raw Normal View History

; RUN: opt < %s -basicaa -memcpyopt -dse -S | FileCheck %s
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
target triple = "i686-apple-darwin9"
%0 = type { x86_fp80, x86_fp80 }
%1 = type { i32, i32 }
define void @test1(%0* sret %agg.result, x86_fp80 %z.0, x86_fp80 %z.1) nounwind {
entry:
%tmp2 = alloca %0
%memtmp = alloca %0, align 16
%tmp5 = fsub x86_fp80 0xK80000000000000000000, %z.1
call void @ccoshl(%0* sret %memtmp, x86_fp80 %tmp5, x86_fp80 %z.0) nounwind
%tmp219 = bitcast %0* %tmp2 to i8*
%memtmp20 = bitcast %0* %memtmp to i8*
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp219, i8* %memtmp20, i32 32, i32 16, i1 false)
%agg.result21 = bitcast %0* %agg.result to i8*
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %agg.result21, i8* %tmp219, i32 32, i32 16, i1 false)
ret void
; Check that one of the memcpy's are removed.
;; FIXME: PR 8643 We should be able to eliminate the last memcpy here.
; CHECK-LABEL: @test1(
; CHECK: call void @ccoshl
; CHECK: call void @llvm.memcpy
; CHECK-NOT: llvm.memcpy
; CHECK: ret void
}
declare void @ccoshl(%0* nocapture sret, x86_fp80, x86_fp80) nounwind
; The intermediate alloca and one of the memcpy's should be eliminated, the
; other should be related with a memmove.
define void @test2(i8* %P, i8* %Q) nounwind {
%memtmp = alloca %0, align 16
%R = bitcast %0* %memtmp to i8*
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %R, i8* %P, i32 32, i32 16, i1 false)
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %Q, i8* %R, i32 32, i32 16, i1 false)
ret void
; CHECK-LABEL: @test2(
; CHECK-NEXT: call void @llvm.memmove{{.*}}(i8* %Q, i8* %P
; CHECK-NEXT: ret void
}
@x = external global %0
define void @test3(%0* noalias sret %agg.result) nounwind {
%x.0 = alloca %0
%x.01 = bitcast %0* %x.0 to i8*
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %x.01, i8* bitcast (%0* @x to i8*), i32 32, i32 16, i1 false)
%agg.result2 = bitcast %0* %agg.result to i8*
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %agg.result2, i8* %x.01, i32 32, i32 16, i1 false)
ret void
; CHECK-LABEL: @test3(
; CHECK-NEXT: %agg.result1 = bitcast
; CHECK-NEXT: call void @llvm.memcpy
; CHECK-NEXT: ret void
}
; PR8644
define void @test4(i8 *%P) {
%A = alloca %1
%a = bitcast %1* %A to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %P, i64 8, i32 4, i1 false)
call void @test4a(i8* align 1 byval %a)
ret void
; CHECK-LABEL: @test4(
; CHECK-NEXT: call void @test4a(
}
declare void @test4a(i8* align 1 byval)
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
declare void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* nocapture, i8 addrspace(1)* nocapture, i64, i32, i1) nounwind
%struct.S = type { i128, [4 x i8]}
@sS = external global %struct.S, align 16
declare void @test5a(%struct.S* align 16 byval) nounwind ssp
; rdar://8713376 - This memcpy can't be eliminated.
define i32 @test5(i32 %x) nounwind ssp {
entry:
%y = alloca %struct.S, align 16
%tmp = bitcast %struct.S* %y to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %tmp, i8* bitcast (%struct.S* @sS to i8*), i64 32, i32 16, i1 false)
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%a = getelementptr %struct.S, %struct.S* %y, i64 0, i32 1, i64 0
store i8 4, i8* %a
call void @test5a(%struct.S* align 16 byval %y)
ret i32 0
; CHECK-LABEL: @test5(
; CHECK: store i8 4
; CHECK: call void @test5a(%struct.S* byval align 16 %y)
}
;; Noop memcpy should be zapped.
define void @test6(i8 *%P) {
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %P, i64 8, i32 4, i1 false)
ret void
; CHECK-LABEL: @test6(
; CHECK-NEXT: ret void
}
; PR9794 - Should forward memcpy into byval argument even though the memcpy
; isn't itself 8 byte aligned.
%struct.p = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 }
define i32 @test7(%struct.p* nocapture align 8 byval %q) nounwind ssp {
entry:
%agg.tmp = alloca %struct.p, align 4
%tmp = bitcast %struct.p* %agg.tmp to i8*
%tmp1 = bitcast %struct.p* %q to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %tmp, i8* %tmp1, i64 48, i32 4, i1 false)
%call = call i32 @g(%struct.p* align 8 byval %agg.tmp) nounwind
ret i32 %call
; CHECK-LABEL: @test7(
; CHECK: call i32 @g(%struct.p* byval align 8 %q) [[NUW:#[0-9]+]]
}
declare i32 @g(%struct.p* align 8 byval)
declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
; PR11142 - When looking for a memcpy-memcpy dependency, don't get stuck on
; instructions between the memcpy's that only affect the destination pointer.
@test8.str = internal constant [7 x i8] c"ABCDEF\00"
define void @test8() {
; CHECK: test8
; CHECK-NOT: memcpy
%A = tail call i8* @malloc(i32 10)
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%B = getelementptr inbounds i8, i8* %A, i64 2
tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %B, i8* getelementptr inbounds ([7 x i8], [7 x i8]* @test8.str, i64 0, i64 0), i32 7, i32 1, i1 false)
%C = tail call i8* @malloc(i32 10)
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%D = getelementptr inbounds i8, i8* %C, i64 2
tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %D, i8* %B, i32 7, i32 1, i1 false)
ret void
; CHECK: ret void
}
declare noalias i8* @malloc(i32)
; rdar://11341081
%struct.big = type { [50 x i32] }
define void @test9_addrspacecast() nounwind ssp uwtable {
entry:
; CHECK-LABEL: @test9_addrspacecast(
; CHECK: f1
; CHECK-NOT: memcpy
; CHECK: f2
%b = alloca %struct.big, align 4
%tmp = alloca %struct.big, align 4
call void @f1(%struct.big* sret %tmp)
%0 = addrspacecast %struct.big* %b to i8 addrspace(1)*
%1 = addrspacecast %struct.big* %tmp to i8 addrspace(1)*
call void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* %0, i8 addrspace(1)* %1, i64 200, i32 4, i1 false)
call void @f2(%struct.big* %b)
ret void
}
define void @test9() nounwind ssp uwtable {
entry:
; CHECK: test9
; CHECK: f1
; CHECK-NOT: memcpy
; CHECK: f2
%b = alloca %struct.big, align 4
%tmp = alloca %struct.big, align 4
call void @f1(%struct.big* sret %tmp)
%0 = bitcast %struct.big* %b to i8*
%1 = bitcast %struct.big* %tmp to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 200, i32 4, i1 false)
call void @f2(%struct.big* %b)
ret void
}
; rdar://14073661.
; Test10 triggered assertion when the compiler try to get the size of the
; opaque type of *x, where the x is the formal argument with attribute 'sret'.
%opaque = type opaque
declare void @foo(i32* noalias nocapture)
define void @test10(%opaque* noalias nocapture sret %x, i32 %y) {
%a = alloca i32, align 4
store i32 %y, i32* %a
call void @foo(i32* noalias nocapture %a)
%c = load i32, i32* %a
%d = bitcast %opaque* %x to i32*
store i32 %c, i32* %d
ret void
}
declare void @f1(%struct.big* nocapture sret)
declare void @f2(%struct.big*)
; CHECK: attributes [[NUW]] = { nounwind }
; CHECK: attributes #1 = { argmemonly nounwind }
; CHECK: attributes #2 = { nounwind ssp }
; CHECK: attributes #3 = { nounwind ssp uwtable }